x265 HEVC Encoder

[Forteen88]

Originally Posted by Barough

x265 v3.2+15-04db2bfee5d6 (32 & 64-bit 8/10/12bit Multilib Windows Binaries) (GCC 9.2.0)

Is this version newer than x265-3.2.1+1?

[Barough]

Yes it is, this is compiled from the default branch and not the stable one.

Skickat från min SM-G975F via Tapatalk

[benwaggoner]

Originally Posted by foxyshadis

How is that any different from physical film grain on actual film stock? That x264 and x265 completely neglected it because it was hard to test doesn't mean a little noise generation isn't valuable.

The grain generation is often done in wrong ways that make it much harder to encode than real world grain. I especially see this in 4K and HDR, where grain gets rendered at the pixel level with SDR code values and so comes out way too bright and too fine-grained. I've seen synthetic film grain that made content impossible to encode at even vaguely acceptable quality at 40 Mbps.

We're also seeing silly amounts of film grain in restored rescans of movies. It isn't creative intent to preserve all kinds of noise that wouldn't have been visible on the perf screens and foot lamberts that the creatives actually approved the content on.

[Blue_MiSfit]

I hear you on the remastering, Ben.

Home Alone is an astounding example of this.

[benwaggoner]

Originally Posted by nghiabeo20

I encode my clip in 5 parts, using -ss and -t to seek to cut position. Now mkvmerge refuses to merge the file, warning that "codec private data doesn't match". After checking the encoding settings, I found that only the --numa-pool change in the first 2 file. How can I edit SPS/PPS data to fool mkvmerge into merging my files? Thanks!

Also maybe try --hrd-concat if you are using HRD?

[Stereodude]

Watch the 4k77 UHD release of Star Wars if you want to see overdone grain. MCTD is calling its name when I build my new Zen 2 system in the next month or two.

[markiemarcus]

Originally Posted by Blue_MiSfit

I hear you on the remastering, Ben.

Home Alone is an astounding example of this.

Something definitely looked off with Home Alone. The pitch of the grain was both far too fine and too bright; it didn't gel at all with the image.

I must say though, I like the way the UHD release of Predator looks and it's pretty damn grainy. It looks right to me. Was shaky film stock from the outset IIRC.

[RanmaCanada]

The original Predator, in my view, is the PERFECT example of what a transfer and remaster should look like. I'm old enough to have seen it in the cinema, and watching that 4k copy was everything I remembered. And we all know how we view the past with rose coloured glasses.

[LigH]

Film grain ... I witnessed an attempt to preserve a sensible amount of it while trying to remove the majority that only distracts the MPEG-2 encoder of a "Kinowelt" DVD production. Source: Die Feuerzangenbowle (1944) – too much denoising, and the trees on the schoolyard look like forged of concrete.

[markiemarcus]

Originally Posted by RanmaCanada

The original Predator, in my view, is the PERFECT example of what a transfer and remaster should look like. I'm old enough to have seen it in the cinema, and watching that 4k copy was everything I remembered. And we all know how we view the past with rose coloured glasses.

Was quite a treat to watch it again I must say. I wonder what the difference in process was, because it obviously wasn't an easy source to work with.

It's a nice encode too. I've noticed that a number of grainy UHD Blu-rays have issues from time-to-time in that regard. There are a handful of scenes in both Blade Runner and The Fog that completely fall apart. Seems inexcusable given how much space there is to play with. In the case of The Fog, the bit rate actually plummets in the affected scenes.

[benwaggoner]

Originally Posted by markiemarcus

Was quite a treat to watch it again I must say. I wonder what the difference in process was, because it obviously wasn't an easy source to work with.

It's a nice encode too. I've noticed that a number of grainy UHD Blu-rays have issues from time-to-time in that regard. There are a handful of scenes in both Blade Runner and The Fog that completely fall apart. Seems inexcusable given how much space there is to play with. In the case of The Fog, the bit rate actually plummets in the affected scenes.

Grain is essentially uncompressible random noise. It's random spatially and temporally, includes chroma, and is typically very high frequency. So it can use the vast majority of bits encoding a frame, is sensitive to QP differences between I, P, B, and b frames, and is otherwise a worst-case nightmare.

While encoding 4K content might take only 2x the bitrate as 1080p, doing grain is much more linear. 80 Mbps 4K is only twice that of Blu-ray 1080p. There are cases of grain that is quite literally impossible to encode without perceptible artifacts. Heck, just encoding a flat gray slide with JUST grain on it can be very challenging.

And that grain, even in lossless, can be quite distracting to a viewer. There's certainly way more fine detail grain on the Blade Runner Blu-ray than Ridley Scott ever saw when he was working on it, due to perf screens, dimmer projectors, etcetera. And in HDR, it's also way brighter grain than would have been seen before. One risk with SMPTE 2100 HDR is that any filters that assume gamma instead of linear light will do weird high contrast things in high luma, since the code values have much bigger jumps in luma than 709 gamma had. Conversely, low luma might not get enough filtering if processed assuming 709.

[jlpsvk]

i have a strange problem backuping my Hobbs and Shaw UHD... It has HDR10+ and when i extract HDR10+ metadata (tried all version of hdr10plus_parser, extracting from raw HEVC stream) and use it for encode mediainfo is like this...

source:

Format                                   : HEVC
Format/Info                              : High Efficiency Video Coding
Format profile                           : Main 10@L5.1@High
HDR format                               : SMPTE ST 2094 App 4, Version 1, HDR10+ Profile B compatible
Width                                    : 3 840 pixels
Height                                   : 2 160 pixels
Display aspect ratio                     : 16:9
Frame rate                               : 23.976 (24000/1001) FPS
Color space                              : YUV
Chroma subsampling                       : 4:2:0 (Type 2)
Bit depth                                : 10 bits
Color range                              : Limited
Color primaries                          : BT.2020
Transfer characteristics                 : PQ
Matrix coefficients                      : BT.2020 non-constant
Mastering display color primaries        : Display P3
Mastering display luminance              : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level              : 1000 cd/m2
Maximum Frame-Average Light Level        : 419 cd/m2

encoded mediainfo (no info about HDR10, HDR10+) weird, json metadata file us about 139MB:

Format                                   : HEVC
Format/Info                              : High Efficiency Video Coding
Format profile                           : Main 10@L5.1@High
HDR format                               : SMPTE ST 2094 App 4, Version 1
Width                                    : 3 840 pixels
Height                                   : 2 160 pixels
Display aspect ratio                     : 16:9
Frame rate                               : 23.976 (24000/1001) FPS
Color space                              : YUV
Chroma subsampling                       : 4:2:0 (Type 2)
Bit depth                                : 10 bits
Color range                              : Limited
Color primaries                          : BT.2020
Transfer characteristics                 : PQ
Matrix coefficients                      : BT.2020 non-constant
Mastering display color primaries        : Display P3
Mastering display luminance              : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level              : 1000 cd/m2
Maximum Frame-Average Light Level        : 419 cd/m2

[mini-moose]

Originally Posted by jlpsvk

i have a strange problem backuping my Hobbs and Shaw UHD... It has HDR10+ and when i extract HDR10+ metadata (tried all version of hdr10plus_parser, extracting from raw HEVC stream) and use it for encode

You should read and post on the hdr10+ thread:
https://forum.doom9.org/showthread.php?t=175947&page=5

I think this happens when the hdr10+ json isn't quite right.

[markiemarcus]

Originally Posted by benwaggoner

Grain is essentially uncompressible random noise. It's random spatially and temporally, includes chroma, and is typically very high frequency. So it can use the vast majority of bits encoding a frame, is sensitive to QP differences between I, P, B, and b frames, and is otherwise a worst-case nightmare.

While encoding 4K content might take only 2x the bitrate as 1080p, doing grain is much more linear. 80 Mbps 4K is only twice that of Blu-ray 1080p. There are cases of grain that is quite literally impossible to encode without perceptible artifacts. Heck, just encoding a flat gray slide with JUST grain on it can be very challenging.

I hadn't really considered the HDR component, that makes sense! Grain compressibility is definitely something I've run into when playing around with x265 on very grainy animation. Assuming semi-transparency as a goal, you're lucky to see a 30-40% reduction at 1080p over x264. That's still pretty good (on the otherwise flat cells the difficulties with grain retention are more obvious). I find live action easier in that regard. You can definitely get there, it just requires some care. I've started leaning towards slightly elevated Psy-rd values (in the range of 2.5 to 3.0), but greatly reduced Psy-RDOQ values (typically below 0.5). I discovered that by accident, but reading through the documentation, it seems that RDOQ by design is less accurate.

With very grainy animated sources, it's quite apparent that RDOQ exaggerates grain coarseness even at the default value of 1.0. It serves a purpose for sure (it's great at maintaining/approximating the motion), I'm just careful with it; it can make a right mess of flat scenes if the grain is intermittent. I've had really good results with CTU 64 and QG 64 in most animated content, grainy or otherwise. I can absolutely see why CTU 64 is the default; I don't find it to cause any difficulties whatsoever with grain retention. Quite the opposite.

The curious thing with The Fog is that the bit rate falls drastically in the troublesome scenes. Seems like an authoring error.

[redbtn]

Originally Posted by markiemarcus

I've started leaning towards slightly elevated Psy-rd values (in the range of 2.5 to 3.0), but greatly reduced Psy-RDOQ values (typically below 0.5).

What about turning psy-rdoq off? It gives worse results than 0.5? And what values you use for grainy movies? I use psy-rd 2.0 psy-rdoq 1.2 for movies, but I'm still not sure about that, still looking for optimal values.

[markiemarcus]

Originally Posted by redbtn

What about turning psy-rdoq off? It gives worse results than 0.5? And what values you use for grainy movies? I use psy-rd 2.0 psy-rdoq 1.2 for movies, but I'm still not sure about that, still looking for optimal values.

You can reduce it to zero, yes, but you can also change RDOQ Level to 0 which produces different results again IIRC. You may not be able to capture all the grain and detail though. It depends; a lot of the time you can get away with it. This is with --no-sao --selective-sao 0 by the way. It looks nice and the grain is certainly uniform, but it doesn't have the same presence as with some RDOQ. Like I said, it serves a purpose. If the content is super grainy, you can probably get away with reducing Aq-strength quite a bit also.

You'll have to experiment; most of my testing was done on grainy animation so I don't know how that plays out on live action. On the grainiest of animated content, my values are around Psy-rd 4.0, Psy-rdoq 0.20, AQ-strength 0.7, Deblocking -4:-1, Aq mode 1, CRF 19.5. You'll also need to get your deblocking strength down to around -4 (though -3 is often sufficient). The other values are somewhere in between the slow and slower preset but with rect disabled.

Edit: If the source has extremely fine grain, more like dither, I've seen better results from disabling RDOQ entirely (Level 0) and pushing Psy-rd as high as 4.50. RDOQ just seems to make a mess; I don't know why.
Edit 2: --selective-sao 1 is generally speaking safe enough for grain retention.

[Ajvar]

Originally Posted by benwaggoner

Grain is essentially uncompressible random noise. It's random spatially and temporally, includes chroma, and is typically very high frequency. So it can use the vast majority of bits encoding a frame, is sensitive to QP differences between I, P, B, and b frames, and is otherwise a worst-case nightmare.

While encoding 4K content might take only 2x the bitrate as 1080p, doing grain is much more linear. 80 Mbps 4K is only twice that of Blu-ray 1080p. There are cases of grain that is quite literally impossible to encode without perceptible artifacts. Heck, just encoding a flat gray slide with JUST grain on it can be very challenging.

And that grain, even in lossless, can be quite distracting to a viewer. There's certainly way more fine detail grain on the Blade Runner Blu-ray than Ridley Scott ever saw when he was working on it, due to perf screens, dimmer projectors, etcetera.

I remember years ago when x265 was at stage of v1.6 - 1.7 people complained about default denoising of HEVC. I remember a grey wall all in terrible grain and HEVC encode then made grain disappear like it never existed and without any artificial look. People complained at x265 encoder exactly for this. Just saying.

[Selur]

What is going on here.
Using 2pass encoding:

1st pass:

i:\Hybrid\64bit\ffmpeg.exe -y -loglevel fatal -noautorotate -nostdin -threads 8 -i "F:\TestClips&Co\files\test.avi" -map 0:0 -an -sn -vf  scale=out_range=pc -pix_fmt yuv420p10le -strict -1 -vsync 0  -f yuv4mpegpipe - | i:\Hybrid\64bit\x265.exe --preset slow --log-level 2 --input - --output-depth 10 --y4m --profile main10 --ctu 32 --no-hme --merange 26 --no-rect --max-merge 2 --tskip --no-open-gop --opt-ref-list-length-pps --bframes 5 --rc-lookahead 40 --pass 1 --slow-firstpass --bitrate 4000 --opt-qp-pps --cbqpoffs -2 --crqpoffs -2 --fast-intra --psy-rd 2.20 --psy-rdoq 1.20 --splitrd-skip --deblock=-2:-2 --selective-sao 2 --limit-sao --no-repeat-headers --range limited --colormatrix bt709 --stats "E:\Temp\result_new.stats" --output "E:\Temp\result_new.265"
y4m  [info]: 640x352 fps 25/1 i420p10 sar 1:1 unknown frame count
raw  [info]: output file: E:\Temp\result_new.265
x265 [info]: HEVC encoder version 3.2+15-04db2bfee5d6
x265 [info]: build info [Windows][GCC 9.2.0][64 bit] 10bit
x265 [info]: using cpu capabilities: MMX2 SSE2Fast LZCNT SSSE3 SSE4.2 AVX FMA3 BMI2 AVX2
x265 [info]: Main 10 profile, Level-3 (Main tier)
x265 [info]: Thread pool created using 16 threads
x265 [info]: Slices                              : 1
x265 [info]: frame threads / pool features       : 4 / wpp(11 rows)
x265 [warning]: Source height < 720p; disabling lookahead-slices
x265 [info]: Coding QT: max CU size, min CU size : 32 / 8
x265 [info]: Residual QT: max TU size, max depth : 32 / 1 inter / 1 intra
x265 [info]: ME / range / subpel / merge         : star / 26 / 3 / 2
x265 [info]: Keyframe min / max / scenecut / bias: 25 / 250 / 40 / 5.00
x265 [info]: Cb/Cr QP Offset                     : -2 / -2
x265 [info]: Lookahead / bframes / badapt        : 40 / 5 / 2
x265 [info]: b-pyramid / weightp / weightb       : 1 / 1 / 0
x265 [info]: References / ref-limit  cu / depth  : 4 / on / on
x265 [info]: AQ: mode / str / qg-size / cu-tree  : 2 / 1.0 / 32 / 1
x265 [info]: Rate Control / qCompress            : ABR-4000 kbps / 0.60
x265 [info]: tools: limit-modes rd=4 psy-rd=2.20 rdoq=2 psy-rdoq=1.20 rskip
x265 [info]: tools: splitrd-skip tskip signhide tmvp fast-intra
x265 [info]: tools: strong-intra-smoothing deblock(tC=-2:B=-2) sao
x265 [info]: tools: selective-sao stats-write
x265 [info]: frame I:      3, Avg QP:6.61  kb/s: 6660.13
x265 [info]: frame P:     82, Avg QP:4.37  kb/s: 9294.05
x265 [info]: frame B:    344, Avg QP:8.83  kb/s: 2552.47
x265 [info]: Weighted P-Frames: Y:2.4% UV:1.2%
x265 [info]: consecutive B-frames: 4.7% 0.0% 18.8% 7.1% 1.2% 68.2%

encoded 429 frames in 11.25s (38.14 fps), 3869.80 kb/s, Avg QP:7.96

and then the second pass fails:

i:\Hybrid\64bit\ffmpeg.exe -y -loglevel fatal -noautorotate -nostdin -threads 8 -i "F:\TestClips&Co\files\test.avi" -map 0:0 -an -sn -vf  scale=out_range=pc -pix_fmt yuv420p10le -strict -1 -vsync 0  -f yuv4mpegpipe - | i:\Hybrid\64bit\x265.exe --preset slow --log-level 2 --input - --output-depth 10 --y4m --profile main10 --ctu 32 --no-hme --merange 26 --no-rect --max-merge 2 --tskip --no-open-gop --opt-ref-list-length-pps --bframes 5 --rc-lookahead 40 --pass 2 --bitrate 4000 --opt-qp-pps --cbqpoffs -2 --crqpoffs -2 --fast-intra --psy-rd 2.20 --psy-rdoq 1.20 --splitrd-skip --deblock=-2:-2 --selective-sao 2 --limit-sao --no-repeat-headers --range limited --colormatrix bt709 --stats "E:\Temp\result_new.stats" --no-dynamic-refine --refine-ctu-distortion 0 --output "E:\Output\result_new.265"

y4m [info]: 640x352 fps 25/1 i420p10 sar 1:1 unknown frame count
raw [info]: output file: E:\Output\result_new.265
x265 [info]: HEVC encoder version 3.2+15-04db2bfee5d6
x265 [info]: build info [Windows][GCC 9.2.0][64 bit] 10bit
x265 [info]: using cpu capabilities: MMX2 SSE2Fast LZCNT SSSE3 SSE4.2 AVX FMA3 BMI2 AVX2
x265 [info]: Main 10 profile, Level-3 (Main tier)
x265 [info]: Thread pool created using 16 threads
x265 [info]: Slices : 1
x265 [info]: frame threads / pool features : 4 / wpp(11 rows)
x265 [warning]: Source height < 720p; disabling lookahead-slices
x265 [error]: different scenecut setting than first pass (40 vs 40)
x265 [error]: failed to open encoder[/code]

x265 [error]: different scenecut setting than first pass (40 vs 40)
40 isn't 40 ?!?

When using a FullHD samples the line with the warning disappears, but the error remains.

Tried multiple sample, with different characteristics, error stays the same, so this does not seem to be source related.

=> Does anyone see that I did wrong, or is this a bug?

Cu Selur

[Selur]

Seems like 2pass is broken or I'm really overlooking something.
Even using a simple settings I end up with the 'different scenecut setting than first pass (40 vs 40)'-error.

I:\Hybrid\64bit>ffmpeg -y -loglevel fatal -noautorotate -nostdin -threads 8 -i "F:\TestClips&Co\files\test.avi" -map 0:0 -an -sn -vf zscale=rangein=tv:range=tv -pix_fmt yuv420p10le -strict -1 -vsync 0  -f yuv4mpegpipe - | x265 --preset medium --input - --output-depth 10 --y4m --profile main10 --pass 1 --slow-firstpass --bitrate 1500  --range limited --colormatrix bt470bg --stats "E:\Temp\test_19_40_50_2910_01.stats" --output "E:\Temp\test.265"
y4m  [info]: 640x352 fps 25/1 i420p10 sar 1:1 unknown frame count
raw  [info]: output file: E:\Temp\test.265
x265 [info]: HEVC encoder version 3.2+15-04db2bfee5d6
x265 [info]: build info [Windows][GCC 9.2.0][64 bit] 10bit
x265 [info]: using cpu capabilities: MMX2 SSE2Fast LZCNT SSSE3 SSE4.2 AVX FMA3 BMI2 AVX2
x265 [info]: Main 10 profile, Level-2.1 (Main tier)
x265 [info]: Thread pool created using 16 threads
x265 [info]: Slices                              : 1
x265 [info]: frame threads / pool features       : 4 / wpp(6 rows)
x265 [warning]: Source height < 720p; disabling lookahead-slices
x265 [info]: Coding QT: max CU size, min CU size : 64 / 8
x265 [info]: Residual QT: max TU size, max depth : 32 / 1 inter / 1 intra
x265 [info]: ME / range / subpel / merge         : hex / 57 / 2 / 3
x265 [info]: Keyframe min / max / scenecut / bias: 25 / 250 / 40 / 5.00
x265 [info]: Lookahead / bframes / badapt        : 20 / 4 / 2
x265 [info]: b-pyramid / weightp / weightb       : 1 / 1 / 0
x265 [info]: References / ref-limit  cu / depth  : 3 / off / on
x265 [info]: AQ: mode / str / qg-size / cu-tree  : 2 / 1.0 / 32 / 1
x265 [info]: Rate Control / qCompress            : ABR-1500 kbps / 0.60
x265 [info]: tools: rd=3 psy-rd=2.00 early-skip rskip signhide tmvp b-intra
x265 [info]: tools: strong-intra-smoothing deblock sao stats-write
x265 [info]: frame I:      3, Avg QP:13.44  kb/s: 3999.20
x265 [info]: frame P:    119, Avg QP:10.20  kb/s: 3036.26
x265 [info]: frame B:    307, Avg QP:14.84  kb/s: 782.98
x265 [info]: Weighted P-Frames: Y:0.8% UV:0.8%
x265 [info]: consecutive B-frames: 2.5% 0.0% 58.2% 22.1% 17.2%

encoded 429 frames in 10.47s (40.97 fps), 1430.51 kb/s, Avg QP:13.54

I:\Hybrid\64bit>ffmpeg -y -loglevel fatal -noautorotate -nostdin -threads 8 -i "F:\TestClips&Co\files\test.avi" -map 0:0 -an -sn -vf zscale=rangein=tv:range=tv -pix_fmt yuv420p10le -strict -1 -vsync 0  -f yuv4mpegpipe - | x265 --preset medium --input - --output-depth 10 --y4m --profile main10 --pass 2 --bitrate 1500  --range limited --colormatrix bt470bg --stats "E:\Temp\test_19_40_50_2910_01.stats" --output "E:\Temp\19_40_50_2910_03.265"
y4m  [info]: 640x352 fps 25/1 i420p10 sar 1:1 unknown frame count
raw  [info]: output file: E:\Temp\19_40_50_2910_03.265
x265 [info]: HEVC encoder version 3.2+15-04db2bfee5d6
x265 [info]: build info [Windows][GCC 9.2.0][64 bit] 10bit
x265 [info]: using cpu capabilities: MMX2 SSE2Fast LZCNT SSSE3 SSE4.2 AVX FMA3 BMI2 AVX2
x265 [info]: Main 10 profile, Level-2.1 (Main tier)
x265 [info]: Thread pool created using 16 threads
x265 [info]: Slices                              : 1
x265 [info]: frame threads / pool features       : 4 / wpp(6 rows)
x265 [warning]: Source height < 720p; disabling lookahead-slices
x265 [error]: different scenecut setting than first pass (40 vs 40)
x265 [error]: failed to open encoder

[redbtn]

Originally Posted by Selur

Seems like 2pass is broken or I'm really overlooking something.
Even using a simple settings I end up with the 'different scenecut setting than first pass (40 vs 40)'-error.

http://forum.doom9.org/showthread.php?p=1890848
Possible the same issue.